Standing and sitting active and passive waist training method and system
By using an LSTM model to assess the patient's fatigue propensity index and dynamic safety threshold, and combining it with instantaneous power gradients, a safe mode switching for lumbar rehabilitation training equipment was achieved. This solved the safety hazards caused by computational delays and improved training safety and real-time performance.
Patent Information
- Application Number
- CN202511474214.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-15
- Publication Date
- 2026-02-17
AI Technical Summary
Existing lumbar rehabilitation training equipment has a calculation delay when judging the patient's fatigue state, which leads to untimely mode switching and may cause secondary injuries such as muscle strain or spasm.
The LSTM model is used to assess patients’ fatigue propensity index. Combined with dynamic safety thresholds and instantaneous power gradients, timely switching from active to passive modes is achieved. By constructing a hybrid safety monitoring strategy, which combines macro-trend prediction and micro-instantaneous judgment, the safety threshold is dynamically adjusted to ensure safety.
This allows for timely responses to patient fatigue without sacrificing accuracy in judgment, thus avoiding secondary injury and improving the safety and real-time nature of rehabilitation training.
Smart Images
Figure CN121534367A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of rehabilitation training technology. More specifically, this invention relates to a method and system for standing and sitting active and passive lumbar training. Background Technology
[0002] Lower back rehabilitation training equipment typically includes active and passive training modes. In active mode, the patient actively exerts force to resist the resistance applied by the equipment to exercise the lower back muscles; in passive mode, the equipment guides the patient's lower back to perform relaxing or stretching movements along a preset trajectory.
[0003] To improve rehabilitation outcomes and safety, modern devices aim to intelligently switch from active to passive modes based on the patient's real-time fatigue status. To achieve this, Long Short-Term Memory (LSTM) network algorithms are typically employed. LSTM models can process time-series data, predicting fatigue levels by learning trends in multi-dimensional data such as torque, speed, and angle during training, thus enabling the switching between active and passive modes.
[0004] However, the above algorithm has an inherent limitation: in order to ensure the accuracy of prediction, LSTM models are usually complex in structure and need to analyze data within a long time window, which leads to high computational resource consumption and unavoidable computational delay.
[0005] In rehabilitation training, a scenario with extremely high safety requirements, this delay means there is a dangerous "reaction time difference" between the system's determination that the patient has reached their fatigue limit and the actual execution of the motor command to switch from active to passive mode. During this period, if the patient, already in a state of exhaustion, continues to try to resist the resistance, it is highly likely to cause secondary injuries such as muscle strain or spasm, which contradicts the fundamental purpose of rehabilitation training.
[0006] Therefore, how to eliminate the safety hazards caused by calculation delays while ensuring the accuracy of fatigue assessment, so as to achieve the optimal state of rehabilitation training, is a problem that current technology urgently needs to solve. Summary of the Invention The purpose of this invention is to propose a standing and sitting active and passive lumbar training method and system to solve the problem in the prior art of how to achieve appropriate switching between execution modes while ensuring the accuracy of fatigue judgment, so as to achieve the best rehabilitation training for users; to this end, this invention provides solutions in the following two aspects.
[0007] In a first aspect, the present invention provides a standing and sitting active and passive lumbar training method, including an active mode and a passive mode, the method further comprising: Acquire multidimensional time-series data of the user in active mode, the multidimensional time-series data including real-time torque, motion angle and angular velocity; Based on a pre-trained LSTM model, the series of the multidimensional time series data within a set time period is processed to obtain a fatigue tendency index. A dynamic safety threshold is determined, the value of which is positively correlated with the set base threshold and positively correlated with the fatigue tendency index; Calculate the instantaneous power gradient, which is the ratio of the difference in instantaneous output power of the user at adjacent sampling times to the sampling time interval; In response to the instantaneous power gradient being less than the dynamic safety threshold, the training mode is switched from active mode to passive mode.
[0008] By constructing an LSTM model to assess the user's fatigue state at various time points, and determining the corresponding dynamic safety threshold based on the assessed fatigue state, this dynamic safety threshold is then compared with the user's instantaneous operational changes during training. This separates complex trend analysis from simple real-time triggering. Ultimately, a switching decision is determined, enabling timely response to dangerous situations, fundamentally solving the computational latency problem, maximizing patient training safety, and effectively avoiding secondary injury.
[0009] Optionally, the training process of the LSTM model includes: Acquire training data, which includes historical exercise data and corresponding fatigue scores of multiple users; the fatigue scores are obtained based on the analysis of physiological data of users collected during the training process. The LSTM model is trained using the training data to obtain a trained LSTM model.
[0010] Optionally, the physiological data includes electromyography (EMG) signals and / or blood oxygen saturation.
[0011] Optionally, the dynamic security threshold is calculated as follows: ; Where Tsafe is the dynamic safety threshold; Tbase is the basic threshold; k is a set positive real number; Pfatigue is the fatigue tendency index; and e is a natural constant.
[0012] The aforementioned dynamic adjustment of the safety threshold enables non-linear and intelligent regulation of the safety threshold. Compared to using a fixed threshold or a linearly changing threshold, this formula allows the system to tolerate larger power fluctuations when the patient is energetic (low fatigue propensity index), avoiding false triggering; while when the patient is nearing their fatigue limit (high fatigue propensity index), the safety threshold becomes exponentially more stringent and sensitive. This adaptive adjustment method allows the safety monitoring strategy to more accurately match the individual patient's state, achieving highly personalized and intelligent safety assurance.
[0013] Optionally, the baseline threshold is obtained by performing an initial calibration test on healthy subjects, corresponding to the maximum acceptable power decrease rate of the system under completely fatigue-free conditions.
[0014] Optionally, the method for calculating the instantaneous power gradient includes: ; Where Gpower(t) is the instantaneous power gradient at the current time t, W(t) is the instantaneous output power at the current time t, W(t-Δt) is the instantaneous output power at the previous sampling time t-Δt, and Δt is the sampling time interval.
[0015] Optionally, the method for calculating the instantaneous power gradient further includes: The instantaneous power gradient and short-term power gradient are obtained; the short-term power gradient is obtained by setting a sliding window and calculating the power change within the sliding window. The instantaneous power gradient and the short-term power gradient are weighted and summed to obtain the final instantaneous power gradient.
[0016] Optionally, it also includes the steps of performing noise reduction filtering and normalization processing on the acquired multidimensional motion data to obtain standardized multidimensional time series data.
[0017] Optionally, switching the training mode from active to passive mode also includes sending prompts to the therapist.
[0018] In the second aspect, a standing and sitting active and passive lumbar training system includes: processor; The memory stores computer instructions for a standing and sitting active and passive lumbar training method, which, when executed by the processor, causes the system to perform the aforementioned standing and sitting active and passive lumbar training method.
[0019] The beneficial effects of this invention are as follows: This invention proposes a hybrid safety monitoring strategy combining "macro-trend prediction" and "micro-instantaneous judgment." Specifically, it utilizes an LSTM model to assess the user's long-term fatigue trend and dynamically adjusts a safety threshold accordingly. A simple instantaneous power gradient index is then compared in real-time with this dynamic threshold to trigger mode switching. This method completely solves the response delay problem caused by existing technologies relying solely on complex models, achieving an organic combination of high-precision prediction and high real-time response. Thus, without sacrificing judgment accuracy, it ensures that a protective mechanism is immediately triggered when the patient experiences instantaneous exhaustion, greatly improving the safety of rehabilitation training. Attached Figure Description
[0020] Figure 1 A simplified schematic diagram of a waist training system is shown. Figure 2 This schematically illustrates a flowchart of the steps of a standing and sitting active and passive waist training method in this embodiment; Figure 3 This schematic diagram illustrates a structural block diagram of a standing and sitting active and passive waist training system according to this embodiment; Attached diagram descriptions: 1. Display screen; 2. First handrail; 3. Second handrail; 4. Protective straps; 5. Dynamic balance plate; 6. Rotating belt. Detailed Implementation
[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. As more and more people suffer from lower back pain, lower back training systems have emerged to address this issue. Users can train standing or sitting, and the system includes two modes: active mode and passive mode. Users can switch between active and passive assisted training modes to improve lower back pain.
[0022] like Figure 1 As shown, the waist training system includes a display 1, a first handrail 2, a second handrail 3, a protective strap 4, a dynamic balance board 5, and a rotating belt 6.
[0023] The monitor is used to display the current user's training duration and number of rotations, etc.
[0024] The dynamic balance plate 5 is mounted on the gravity plate, and a motor is installed below the gravity plate to assist the rotation of the dynamic balance plate via a rotating belt. A gyroscope is installed below the dynamic balance plate to measure angles and angular velocities. A torque sensor is also installed at the rotating belt to measure real-time torque.
[0025] The lumbar training system in this embodiment is also equipped with a seat ( Figure 1(Not shown in the image), the seat is used for training when the current user is sitting; while the protective strap 4 is used for training when the current user is standing; whether to train sitting or standing can be determined based on the current user's actual situation.
[0026] In active mode, the tilt angle of the dynamic balance board 5 is set according to the user's physical condition, and the user begins to rotate actively. In passive mode, the patient fastens the protective straps 4, holds the handrails, and undergoes passive assisted training according to the rotation plan. For example, the patient's waist rotation angle and balance board tilt angle during active training are used as the initial values to begin passive training, and the waist rotation angle and balance board tilt angle are appropriately increased to provide passive training for the patient.
[0027] However, when performing lumbar training, if the patient's fatigue level cannot be accurately assessed, the timing of the transition between active and passive modes may indeed be inappropriate, leading to safety and efficacy issues.
[0028] Therefore, based on the above problems, the present invention provides a standing and sitting active and passive lumbar training method and device, which can achieve appropriate switching between execution modes while ensuring the accuracy of fatigue judgment, so that the user's rehabilitation training can achieve the best effect.
[0029] like Figure 2 As shown in this embodiment, a standing and sitting active and passive lumbar training method includes the following steps: Step S1: Obtain the multidimensional time series data of the current user.
[0030] When a patient is performing active lumbar training, the torque sensor and gyroscope set on the lumbar training system synchronously collect the real-time torque F(t), the real-time motion angle θ(t) of the lumbar joint and the real-time angular velocity ω(t) of the current user at each moment at a preset sampling frequency (e.g., 100Hz).
[0031] The acquired raw data sequences are processed through noise reduction filtering and normalization to form standardized multidimensional time series data, providing high-quality data input for subsequent analysis.
[0032] Among these methods, normalization can be achieved using the maximum-minimum normalization method. Noise reduction filtering can employ mean filtering, median filtering, etc.
[0033] Step S2: Calculate the fatigue tendency index of the current user based on the LSTM model; and determine the dynamic safety threshold based on the fatigue tendency index.
[0034] Specifically, the preprocessed multidimensional time series data is input into a pre-trained LSTM model.
[0035] The LSTM (Long Short-Term Memory) model is a special type of recurrent neural network (RNN) designed to address the vanishing or exploding gradient problems that traditional RNNs encounter when processing long sequences of data. In this embodiment, the LSTM model does not directly trigger mode switching, but rather assesses the long-term trend of the patient's macroscopic fatigue state. The model receives data sequences from a relatively long time window (e.g., 10 seconds) and outputs a continuous, standardized fatigue tendency index, Pfatigue, in real time.
[0036] In this embodiment, the fatigue tendency index ranges from [0,1], where 0 represents that the patient is energetic and shows no signs of fatigue; and 1 represents that the patient is at or very close to the physiological fatigue limit. Since Pfatigue reflects a long-term cumulative trend, the slight delay caused by the LSTM model calculation does not affect its effectiveness as a macroscopic state assessment indicator.
[0037] Furthermore, in order to obtain the model's output results more accurately, the fatigue tendency index is also corrected to obtain the corrected fatigue tendency index. Specifically, the product of the model accuracy of the trained LSTM model and the fatigue tendency index at each time step is used as the corrected fatigue tendency index.
[0038] The model accuracy is calculated by dividing the number of correctly predicted samples by the total number of samples.
[0039] The training process of the LSTM model is as follows: First, obtain the training set. The training set includes input data and labels. The input data is historical data from active training by different users, specifically including historical time-series data such as real-time torque F(t), lumbar joint angle θ(t), and angular velocity ω(t).
[0040] The label is the user's fatigue rating during the training process, which is a continuous value ranging from [0, 1].
[0041] The fatigue score is obtained by collecting physiological data from the corresponding user, such as electromyography (EMG) signals, blood oxygen saturation and other dimensions. The fatigue score at different times is obtained by analyzing the amplitude decrease or frequency change of each dimension of data.
[0042] For example, the initial stage of training (without fatigue) is marked as Pfatigue=0; when the EMG signal amplitude drops by 50%, it is marked as Pfatigue=1; and when the EMG signal amplitude drops by 20%, it can be marked as Pfatigue=0.4.
[0043] Of course, as another implementation method, manual annotation can be carried out based on the user's exercise performance and feedback, that is, relying on the experience of the rehabilitation therapist to manually annotate the user during the training process.
[0044] The aforementioned different users cover patient data of different ages, genders, disease types (such as lumbar disc herniation, postoperative rehabilitation, and chronic low back pain) and rehabilitation stages, which enhances the generalization ability of the LSTM model and makes the model more robust.
[0045] Secondly, the LSTM model is trained using the training set to obtain a trained LSTM model.
[0046] Specifically, the training set is input into the network prediction model for training, and the loss value is calculated using a loss function. The parameters of the network prediction model are adjusted using the gradient descent algorithm until the loss value between the output prediction value and the label is less than the threshold or the number of training iterations reaches the set number. At this point, training is stopped, and the trained network prediction model is obtained.
[0047] The loss function mentioned above is the mean squared error loss function. Since the specific training process is based on existing technology, it will not be described in detail here.
[0048] After obtaining the current user's fatigue tendency index at each moment, the corresponding dynamic safety threshold is also determined.
[0049] Specifically, the dynamic safety threshold Tsafe is: ; Where Tsafe is the dynamic safety threshold, a negative value used for comparison with Gpower(t). k is the fatigue sensitivity coefficient, a positive real number used to adjust the degree of change of Tsafe with Pfatigue; Pfatigue is the fatigue susceptibility index.
[0050] Tbase is the basic safety threshold, which represents the maximum rate of power decrease that the system can tolerate when the patient is completely fatigue-free (Pfatigue=0). This value is a negative constant and can be obtained by performing initial calibration tests on healthy subjects. In the above formula, as Pfatigue increases from 0 to 1, the exponent term decreases from 1 towards a positive number close to 0. Since Tbase is negative, this causes Tsafe to gradually increase from this large negative value of Tbase, approaching 0. This means that as the patient's fatigue level increases, the system's tolerance for power decline (i.e., Tsafe) becomes smaller and smaller, and the judgment conditions become more stringent and sensitive.
[0051] The purpose of constructing the dynamic safety threshold mentioned above is that when the patient is energetic (lower Pfatigue), they can withstand larger instantaneous power fluctuations, and the safety threshold should be relatively lenient at this time; while when the patient is fatigued (higher Pfatigue), any small power drop may be a danger signal, and the dynamic safety threshold must become extremely strict at this time.
[0052] Step S3: Obtain the instantaneous power gradient of the current user at each time point; based on the comparison between the dynamic safety threshold and the instantaneous power gradient, switch the training mode of the current user in real time.
[0053] To achieve real-time safety monitoring, it is necessary to construct an indicator that can characterize the patient's instantaneous exertion state, namely the instantaneous power gradient Gpower.
[0054] Considering that the sharp decline in the patient's exertion during training is the most direct signal that the muscles are about to exhaust, the instantaneous power gradient can sensitively capture this change.
[0055] In one embodiment, the process of obtaining the instantaneous power gradient is as follows: First, based on the obtained real-time torque F(t) and real-time angular velocity ω(t), the patient's instantaneous output power W(t) is calculated as follows: W(t) = F(t)•ω(t).
[0056] Secondly, by performing differential calculation on the power signal, the instantaneous power gradient Gpower is obtained, specifically: ; Where Gpower(t) is the instantaneous power gradient at the current time t, reflecting the rate of change of power. W(t) is the instantaneous output power at the current time t; W(t-Δt) is the instantaneous output power at the previous sampling time t-Δt, where Δt is the sampling time interval.
[0057] When Gpower(t) is negative and its absolute value is large, it indicates that the patient's output power is decreasing rapidly, which is a strong warning signal that motor function failure is about to occur.
[0058] Instantaneous power gradient can sensitively reflect the "rate of change in strength." In rehabilitation training, the most direct signal that a patient is about to reach exhaustion is a "sharp drop" in their output power, and the power gradient can capture this danger signal most directly and quickly. Therefore, using this indicator as a real-time judgment basis can predict the risk of motor function failure earlier and more accurately than existing technologies, further enhancing the system's instantaneous response capability and the timeliness of protection.
[0059] Furthermore, in order to more accurately evaluate the training of the current user and reduce misjudgments of noise or short-term fluctuations, a short-term power gradient is introduced in this embodiment. The short-term power gradient and the instantaneous power gradient are weighted and summed to obtain the final instantaneous power gradient.
[0060] Among them, the short-term power gradient calculates the power change within the sliding window by setting a sliding window. The length of the sliding window can be set according to the actual situation, such as 五分钟 (should be "5 minutes" in English).
[0061] The weights for the above weighted summation can be obtained based on experience, such as the short-term instantaneous gradient being 0.3; or an adaptive method can be used to dynamically adjust according to the fatigue tendency index (for example, when the fatigue tendency index is high, increase the weight of the instantaneous gradient).
[0062] However, simply using a fixed Gpower threshold for judgment may be inaccurate because it cannot adapt to the safety standards in different fatigue states. Therefore, the present invention compares the dynamic safety threshold Tsafe with Gpower(t).
[0063] Among them, if Gpower(t) < Tsafe, it is immediately determined that the patient has experienced instantaneous ability failure and there is a safety risk. The system will immediately trigger the protection mechanism, stop the training mode of the current user, and prompt the rehabilitation therapist to switch the training mode from the active mode to the passive mode.
[0064] Since the calculation of Gpower(t) and the comparison with Tsafe are both simple arithmetic operations based on single-sampling data, the calculation time consumption can be ignored. This ensures that the entire process from detecting the danger signal to executing the switching instruction is almost instantaneous, thus completely eliminating the safety hazard caused by the calculation delay of the LSTM model.
[0065] The solution of the present invention, through the accurate evaluation of the long-term fatigue trend by the LSTM model, is cleverly used to dynamically adjust the sensitivity of the real-time safety threshold. This enables the safety monitoring system to "intelligently" adaptively adjust its judgment criteria according to the macroscopic state of the patient, avoiding false triggers due to accidental fluctuations when the patient still has the ability, and ensuring sensitive protection when the patient is truly fatigued, achieving an organic combination of high precision and high real-time performance, thereby correcting the patient's不良坐姿站姿 (should be "bad sitting and standing postures" in English) and improving lumbar muscle strain, etc.
[0066] It should be noted that there was an incorrect Chinese expression "五分钟" in the original text which was translated as "五分钟" inappropriately. It should be "5 minutes". Also, "不良坐姿站姿" was translated as "不良坐姿站姿" wrongly and should be "bad sitting and standing postures".Simultaneously, this method possesses both forward-looking prediction and instantaneous response capabilities. It can not only learn the general fatigue patterns of specific patients (through LSTM), but also react instantly to unpredictable emergencies during training (such as inattention or sudden exhaustion), thus providing highly personalized and adaptive safety guarantees for each patient and significantly improving the performance and user experience of intelligent rehabilitation equipment. This invention also provides a standing-sitting active and passive lumbar training system. For example... Figure 3 As shown, the system includes a processor and a memory, the memory storing computer program instructions, which, when executed by the processor, implement the above-described standing and sitting active and passive waist training method according to the present invention.
[0067] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and therefore will not be described in detail here.
[0068] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented by computer-readable / executable instructions stored or otherwise maintained on such a computer-readable medium.
[0069] In the description of this specification, "multiple" means at least two, such as two, three or more, etc., unless otherwise expressly and specifically defined.
[0070] While various embodiments of the invention have been shown and described in this specification, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention.
Claims
1. A sitting active-passive lumbar training method, comprising an active mode and a passive mode, characterized in that, The method further comprises: acquiring multi-dimensional time series data of the user in the active mode, the multi-dimensional time series data including real-time torque, motion angle and angular velocity; processing a sequence within a set time length of the multi-dimensional time series data based on a pre-trained LSTM model to obtain a fatigue tendency index; determining a dynamic safety threshold, the value of the dynamic safety threshold being positively correlated with a set base threshold and positively correlated with the fatigue tendency index; calculating an instantaneous power gradient, the instantaneous power gradient being a ratio of a difference between instantaneous output powers of the user at adjacent sampling time points and a sampling time interval; switching the training mode from the active mode to the passive mode in response to the instantaneous power gradient being less than the dynamic safety threshold.
2. The method according to claim 1, wherein, The training process of the LSTM model comprises: acquiring training data, the training data including historical motion data of multiple users and corresponding fatigue scores; the fatigue scores being obtained based on physiological data of the users collected in the training process; training the LSTM model using the training data to obtain a trained LSTM model.
3. The method according to claim 2, wherein, The physiological data includes electromyography (EMG) signals and / or blood oxygen saturation.
4. The method according to claim 1, wherein, The calculation method of the dynamic safety threshold is: ； wherein, Tsafe is the dynamic safety threshold; Tbase is the base threshold; k is a set positive real number; Pfatigue is the fatigue tendency index; and e is a natural constant.
5. The method according to claim 4, wherein, The base threshold is obtained by initial calibration testing on healthy subjects, corresponding to an acceptable maximum power decline rate of the system in a completely non-fatigued state.
6. The method according to claim 1, wherein, The calculation method of the instantaneous power gradient comprises: ； wherein, Gpower(t) is the instantaneous power gradient at the current time t; W(t) is the instantaneous output power at the current time t; W(t-Δt) is the instantaneous output power at the previous sampling time t-Δt; and Δt is the sampling time interval.
7. The method as claimed in claim 1, wherein, The calculation method of the instantaneous power gradient further comprises: acquiring the instantaneous power gradient and a short-term power gradient; the short-term power gradient being obtained by setting a sliding window and calculating power changes within the sliding window; weighting and summing the instantaneous power gradient and the short-term power gradient to obtain the final instantaneous power gradient.
8. The method according to claim 7, wherein, Further comprising: a step of performing noise reduction filtering and normalization processing on the acquired multi-dimensional motion data to obtain standardized multi-dimensional time series data.
9. The method as claimed in claim 1, wherein, Switching the training mode from the active mode to the passive mode further comprises sending a prompt message to a physiotherapist.
10. A seated active-passive lumbar training system, comprising: Comprise: a processor; a memory storing computer instructions for a standing and sitting active-passive waist training, when the computer instructions are run by the processor, the system executes a standing and sitting active-passive waist training method according to any one of claims 1-9.